Big Data. George O. Strawn NITRD

Transcription

1 Big Data George O. Strawn NITRD

2 Caveat auditor The opinions expressed in this talk are those of the speaker, not the U.S. government

3 Outline What is Big Data? NITRD's Big Data Research Initiative Big Data in Science and Business

4 What is Big Data? A term applied to data whose size, velocity or complexity is beyond the ability of commonly used software tools to capture, manage, and/or process within a tolerable elapsed time. Volume, velocity, variety, veracity,...

5 What is Big Data, really? "Existence precedes essence" (JP Sartre) BD: Data that precedes its uses Standard IT paradigm: conceive an app; create/collect the data; process the data BD paradigm: create/collect data; conceive apps; process the data

6 Big data requires big computing These days, supercomputers aren't actually bigger: they're broader (thousands of cpu's) Server farms are "loosely coupled" supercomputers (thousands of servers) Big volume data resides on supercomputers or server farms (or at least on clusters)

7 Big Data processing Phase 1 : Ingest Phase 2 : Store Phase 3 : Analyze (three options) Phase 4 : Visualize Phase 5 : Insight/Decide

8 Analyze phase options Distributed Memory Architecture for needlein-haystack applications; e.g., Hadoop Shared-Memory Non-Coherent Architecture Shared-Memory Coherent Architecture for connections-between-hay-in-stack analysis; e.g., DNA de novo assembly

9 The CAP Theorem Consistency, Accessibility, Partitionability Traditional Databases can have all three Big Date can only have two out of three!

10 Big Data includes Data Intensive Science The science community is a driving force for big data (and it's the NITRD focus) But it's often the case that developments in scientific computing have society-wide impact And, science often takes advantage of nonscientific computing developments (eg, gpu's, google's map-reduce)

11 Why now for Big Data? Moore's laws for cpu's, disks, networks, sensors Disk storage cost has gone from 25 cents per byte (IBM 305 Ramac in 1956) to 25 cents per ten gigabytes today. 25 cents per terabyte soon? Sensors: remote sensing, video surveillance, environmental sensing, scientific instruments, etc The Internet: Five billion gigabytes and counting (estimated by Eric Schmidt)

12 NITRD Networking and IT R&D A 21-year-old interagency program to enhance coordination and collaboration of the IT R&D programs of a number of Federal agencies Member agencies Areas of interest

13 NITRD Member Agencies DoC NOAA NIST DoD OSD DARPA AFOSR, ARL, ONR DoE (SCI, NNSA, OE) DHS EPA HHS AHRQ NIH ONC NARA NASA NRO NSA NSF (CISE, OCI)

14 NITRD PCAs (program component areas) Cyber Security and Information Assurance High Confidence Software and Systems High-End Computing Human Computer Interaction and Info Mgmt Large Scale Networking Social, Economic, and Workforce Implications Software Design and Productivity

15 NITRD SSGs (senior steering groups) Cybersecurity Health IT R&D Wireless Spectrum Efficiency CyberPhysical Systems Big Data

16 NITRD s Big Data Initiative Core Technologies Domain Research Data Challenges/Competitions Workforce Development

17 Core Tech I: Collection, Storage and Management of Big Data Data representation, storage and retrieval New parallel data architectures, including clouds Data management policies, including privacy and access Communication and storage devices with extreme capabilities Sustainable economic models for access and preservation

18 Core Tech II: Data Analytics Computational, mathematical, statistical and algorithmic techniques for modeling high dimensional data Learning, inference, prediction and knowledge discovery for large volumes of dynamic data sets Data mining to enable automated hypothesis generation, event correlation and anomaly detection Information infusion of multiple data sources

19 Core Tech III: Data Sharing and Collaboration Tools for distant data sharing, real time visualization and software reuse of complex data sets Cross disciplinary model, information and knowledge sharing Remote operation and real time access to distant data sources and instruments

20 Big Data in Business and Government Business analytics Trends

21 Business Analytics The use of statistical analysis, data mining, forecasting, and optimization to make critical decisions and add value based on customer and operational data. Critical problems are often characterized by massive amounts of data and the need for rapid decisions and high performance computing Eg, modeling customer lifetime value in banks reducing adverse events in health care managing customer relationships in hospitality industry

22 Trend 1: Bigger Data Volume, velocity, variety of big data keep increasing! Storage and compute capacity often less than needed for timely decision Basis for web-based businesses (Google, Facebook, ) Business sectors are leading the way in exploiting data about customers and transactions. Prevalent in pharmaceutical, retail, and financial sectors

23 Trend 2: Unstructured Data 70% of enterprise data is unstructured: images, , documents Text analytics: linguistics, natural language processing, statistics Content categorization, sentiment analysis Text mining: statistical learning applied to a collection of documents Examples: discovery of adverse drug effects from patient notes; identification of fraudulent insurance claims; sentiment analysis based on Facebook posts; early warning from warranty and call center data

24 Trend 3: Distributed Data Terabyte-sized data are spread across multiple computers, and are increasingly held in distributed data stores that are amenable to parallel processing. Extraction into traditional computing environments chokes on data movement Challenge is to co-locate analysis with data Apache Hadoop is now widely used for Big Data applications

25 Trend 4: Distributed Computing Scaling our computational tools, algorithms and thinking: how do we apply parallel programming methods for processing data distributed on thousands of computers How do we acquire specialized programming skills? Where are the data located? What proportion of the work can be done in parallel by nodes? Do we understand the mechanisms that generate Big Data? What are useful models? How do we look further?

26 Big Data in Science Analyzing output from supercomputer simulations (eg, climate simulations) Analyzing instrument (sensor) output Creating databases to support wide collaboration (eg, human genome project) Creating knowledge bases from textual information (eg, Semantic Medline)

27 Scientific Data Analysis Today Scientific data is doubling every year, reaching PBs (CERN is at 22PB today, 10K genomes ~5PB) Data will never again be at a single location Architectures increasingly CPU-heavy, IO-poor Scientists need special features (arrays, GPUs) Most data analysis done on midsize BeoWulf clusters. Universities hitting the power wall Soon we cannot even store the incoming data stream Not scalable, not maintainable

28 LHC tames big data? Produces a petabyte of info per second Saves for processing a petabyte per month This factor of 10**6 reduction in data is possible because i) the LHC is "smart" and ii) there is a "good model" of the data

29 Data in HPC Simulations HPC is an instrument in its own right Largest simulations approach petabytes--from supernovae to turbulence, biology and brain modeling Need public access to the best and latest through interactive numerical laboratories Creates new challenges in: how to move the petabytes of data (high speed networking); how to look at it (render on top of the data, drive remotely) How to interface (virtual sensors, immersive analysis) How to analyze (algorithms, scalable analytics)

30 Common Analysis Patterns Large data aggregates produced, but also need to keep raw data Need for parallelism; heavy use of structured data, multi-d arrays Requests enormously benefit from indexing (eg. rapidly extract small subsets of large data sets) Computations must be close to the data! Very few predefined query patterns Geospatial/locality based searches everywhere Data will never be in one place, and remote joins will not go away No need for transactions, but data scrubbing is crucial

31 Disk Needs Today Disk space, disk space, disk space!!!! Current problems not on Google scale yet: TB easy, 100 TB doable, 300 TB hard For detailed analysis we need to park data for several months Sequential IO bandwidth--if analysis is not sequential for large data set, we cannot do it How to move 100TB within a University? 1Gbps --10 days; 10 Gbps--1 day (but need to share backbone); 100 pound box-- few hours From outside? Dedicated 10Gbps or FedEx

32 Clouds Economy of scale is clear Commercial clouds are too expensive for Big Data--smaller private clouds with special features are emerging May become regional gateways to larger-scale centers The Long Tail of a huge number of small data sets (the integral of the long tail is big) Facebook brings many small, seemingly unrelated data to a single cloud and new value emerges. What is the science equivalent?

33 Science and Big Data Science is increasingly driven by data (large and small) Large data sets are here, COTS solutions are not From hypothesis-driven to data-driven science We need new instruments: microscopes and telescopes for data There is also a problem on the long tail Similar problems present in business and society Data changes not only science, but society A new, Fourth Paradigm of Science is emerging

34 From Bits to Its? After newton, the world consisted of matter in motion After the steam engine came thermodynamics and the world consisted of matter and energy After the computer, perhaps comes a science of information and the world may then consist of matter, energy and information

35 What the future may hold Data intensive science appears to be revolutionary science Data analytics and other big data services are major opportunities for business and government Big Data may also be the basis of new services for people, perhaps as significant as the Web, Google and Facebook